NVIDIA agrees to acquire GPU orchestration software provider Run:ai

Background

Modern AI work runs on GPUs, and GPUs are expensive and in limited supply. A single cluster usually serves many teams, many jobs and many priorities at once, so someone has to decide which job gets which chip, when it starts and how long it runs. Software that does this bookkeeping sits between the people who want compute and the hardware that supplies it.

Run:ai made GPU orchestration software that helps customers schedule, manage and optimize AI compute infrastructure on premises, in the cloud or in hybrid environments^[2]. The company was headquartered in Israel^[2]. Its products were built around Kubernetes — the open-source system widely used to manage containers — and NVIDIA described Run:ai as a Kubernetes-based workload management and orchestration software provider^[1]. Run:ai co-founder and CEO Omri Geller said the company had been a close collaborator with NVIDIA since 2020^[1].

The agreement

On 24 April 2024, NVIDIA announced that it had entered into a definitive agreement to acquire Run:ai^[1]. NVIDIA did not disclose the price; the deal was reported at approximately $700 million^[3][4]. Along with the announcement, NVIDIA said it would continue to invest in the Run:ai product roadmap as part of NVIDIA DGX Cloud^[1].

Review in Europe

The acquisition was examined by the European Commission. The transaction was notified on 15 November 2024, after Italy referred the case under Article 22 of the EU Merger Regulation^[2], a mechanism that lets EU member states ask the Commission to review a transaction. On 20 December 2024 the Commission unconditionally approved the acquisition^[2].

Completion

NVIDIA completed the acquisition on 30 December 2024^[3]. The parties said that the software, which until then worked only with NVIDIA GPUs, would be open-sourced^[3].

Why it matters

The deal was about software rather than chips. NVIDIA is best known for the processors that AI systems are trained on; Run:ai supplied the layer that decides how those processors are shared among jobs and users^[1][2]. Bringing that layer in-house put a widely used piece of AI infrastructure management inside the company whose hardware the infrastructure is built on.

The second notable element is the licence. Software that until the acquisition worked only with NVIDIA GPUs was to be released as open source^[3], a detail the parties confirmed when the transaction closed.

The timeline also shows how a mid-sized software acquisition is processed in Europe. The agreement was announced in April 2024, the transaction was formally notified in mid-November after Italy's referral, the Commission cleared it unconditionally in December, and the deal closed ten days later^[1][2][3]. From announcement to completion, the process ran for roughly eight months.